Akhilesh Kalnoor

Building Robots @SkildAI.

[Akhilesh Kalnoor @ashkalor]

@jaredmejiaa @gschw1nd Credit where credit is due.

[Akhilesh Kalnoor @ashkalor]

@jaredmejiaa ikr!! what crazy times to live in

[Akhilesh Kalnoor @ashkalor]

@deepakpathak Its crazy that the plan was this simple.

[Akhilesh Kalnoor @ashkalor]

@SkildAI

[Akhilesh Kalnoor @ashkalor]

12yr old me
I love transformers

26yr old me
I love transformers

"Attention is all you need"

[Akhilesh Kalnoor @ashkalor]

If y'll didnt see the first video properly, see it again. Putting a plate in a toaster. ICL allows us to really attempt things that were previously not so straightforward.

[Deepak Pathak @deepakpathak]

"Does S1 exhibit physical prompt steerability: different prompts induce distinct behavior in the same environment?"

Yes! Watch S1 follow 4 different video prompt recipes in the same kitchen.

The first one is something far out of distribution -- "putting a plate in a toaster".

[Akhilesh Kalnoor @ashkalor]

Coolest thing about s1 imo, been a fan of it from day 1. Only the real ones know how it feels when the model misses by a few cms and unable to correct itself.

[Deepak Pathak @deepakpathak]

If S1 makes a mistake, it tries again.

For example, the input video prompt here attached the wheel just once. During deployment the wheel didn't align properly, so S1 re-aligned it.

By watching the prompt, S1 understands the goal and improvises to get there:

[Akhilesh Kalnoor @ashkalor]

@KentonVarda LLVM has a good AI policy related to this.
https://t.co/2hfPfMzN8t

[Akhilesh Kalnoor @ashkalor]

Next couple of weeks are going to be insane. Fable 5.1 and Astra are rumoured to be lined up. It doesnt really matter which model outperforms the other. Two labs shipping means the floor moves for everyone.

[Akhilesh Kalnoor @ashkalor]

What if we just... kept pretraining?

[Deepak Pathak @deepakpathak]

We’re scaling S1 as fast as we can. The curve shows no signs of slowing down, steadily going up.

We’ve been working for a while to find this kind of scaling law in robotics.

[Akhilesh Kalnoor @ashkalor]

@gschw1nd @SkildAI Did you not see the demo? It performs well even if the demonstrator messes up. 👀👀

[Akhilesh Kalnoor @ashkalor]

Some great insights into how this was possible here: https://t.co/49kx92IKbv

[Akhilesh Kalnoor @ashkalor]

SKILD S1 !! Every new task usually meant hours of teleop or fine tuning, LLMs escaped this with In-context learning and now its time for robots to do the same.

The parts that I am really fascinated about is how it tries to achieve the goal rather than just copying the demo. No watering can in the scene ? it uses a cup . Glass already full? it tops it off. Demonstrator drops the egg? S1 doesnt!

[Skild AI @SkildAI]

Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning:

[Akhilesh Kalnoor @ashkalor]

@shagunuppls @SkildAI Congrats @shagunuppls. Cant wait to see what you have in store for us 💪💪

[Mitchell Hashimoto @mitchellh]

Mind boggling to me that I can make a thing faster and there's always people that ask "but why?" What kind of mentality is that? The pursuit of excellence does not need justification. Also, I find in so many cases, we can't know the impact of an improvement until we do it.

For example, one I've talked about before: Ghostty's high IO throughput has enabled terminal program (emulator and TUI) fuzzing at a speed thats incomparably fast to prior solutions. This has resulted in upstream patches to resolve issues in popular projects like btop, tmux, and more.

Speed enabled that anecdotally example that lifted the tides of adjacent communities that don't rely on Ghostty technology at all. I didn't predict this.

Make things better because they can be better and let the results naturally play out.

[Akhilesh Kalnoor @ashkalor]

Congrats!! Many more to come 🚀

[Skild AI @SkildAI]

Exciting news! 🎉 Our CEO & Co-founder, Deepak Pathak (@deepakpathak), received the PAMI Young Researcher Award at #CVPR2026 this week.

Among the highest honors in computer vision for early-career researchers, the award recognizes groundbreaking contributions that have a lasting impact on the field of AI.

Congratulations, Deepak!

[Jamon @jamonholmgren]

Agentic coding tip

Pay down tech debt immediately

With agents, there should be no such thing as “tech debt”. An agent should simply pay down every bit of tech debt before presenting you with the “finished work”. Unlike humans, agent time is not very valuable; it can and should continue to work on something until it’s done, and shouldn’t make concessions assuming human constraints.

If an agent tells you that it’s “leaving something for later”, tell it to go finish it first before saying it’s done.

Example prompt (best if it is in your docs):

```
Do not leave any tech debt behind. If you have taken any shortcuts, go back and do them right. This is a hard acceptance criteria that must be completed.
```